Synthesis: from findings to a usable output.

What synthesis is, and what it is not CORE

By this point in Module 4 you have a scoping brief, a set of triangulated claims, and a list of opened-and-confirmed sources. You have the raw material of a research output. What you do not yet have is a research output.

Synthesis is the move that turns a set of triangulated claims into a single document a human reader can use. It has three jobs:

Shape. Decide what order the reader needs the information in. Research is not consumed in the order you found it. The shape of a good brief reflects the reader’s path through the question, not the researcher’s path through the sources.

Selection. Decide what goes in and what stays out. A triangulated set of claims is almost always too much material for the brief. Synthesis picks the load-bearing claims and discards or compresses the rest. A brief that includes every claim you found is a brief that has skipped selection.

Connection. Write the sentences between the claims. This is the prose-craft part of synthesis — the connector sentences, the transitions, the topic sentences, the conclusion. This is also where the most honesty-at-risk happens, because connector prose is exactly where a model’s pull toward confident-sounding writing can smuggle in claims that no source supports.

Synthesis is not summarization. A summary says: “Here are the things I found.” Synthesis says: “Here is the answer to your question, supported by what I found, shaped for the reader you named, with the uncertainty left where it actually lives.” The difference is whether the output is organized around the sources or around the question.

Students routinely ask a research agent to “write up the findings.” The output looks like a research brief and is usually a summary in brief’s clothing — ordered by what the agent retrieved, not by what the reader needs. Synthesis starts from the other direction: from the reader’s question, down into the claims, with the sources as footnote rather than structure.

The three failure modes of synthesis CORE

Before you write a synthesis plan, know what the three common failures look like. You will see all three in your own work and in agent output. Catching them is part of the craft.

1. The false-voice problem. The agent smooths disagreement into confident prose. A set of sources where two studies find an effect and one finds no effect becomes: “Research suggests the effect is modest but real.” That sentence contains a judgment the sources did not collectively make. It sounds authoritative, it is comfortable to read, and it is lying a little bit. The fix: in the synthesis, name the disagreement (“Two recent studies report a modest effect; a third, controlling for district size, does not”). Let the reader feel the real state of knowledge.

2. The fact-free connector. The agent writes a paragraph that is mostly transitional sentences — “This is important because…”, “In light of these findings…”, “It is worth noting that…” — and almost no sourced claims. These paragraphs read smoothly. They say nothing. The fix: every paragraph must carry at least one sourced claim. Connector prose earns its keep by moving between claims, not by filling space.

3. The single-source paragraph passed off as consensus. A paragraph is built on one source, but its phrasing suggests broad agreement. “Experts agree that X,” when the “experts” are one person in one paper. The fix: where a claim rests on one source, name the source inline (“Thompson (2024) argues that X”) rather than using anonymous plural voice.

Each of these is fixable once you see it. The skill is seeing it. Read your own drafts and the agent’s drafts with these three failure modes in mind.

The synthesis plan CORE

You do not hand the agent a pile of source notes and ask for a brief. You hand the agent a synthesis plan, and ask it to draft the brief against the plan.

A synthesis plan is a one-page outline with six parts.

1. The question, copied from your scoping brief.

2. The thesis sentence. One sentence that gives the answer. Not the topic sentence — the answer. “Beast Academy fits our family better than Singapore Math at the 4th-grade level, but the case flips at 7th grade; we should use different curricula for the two students.” Write this before you write the brief. If you cannot write it, your triangulation did not produce a verdict, and you need either more evidence or a narrower question.

3. The structure, as a list of section headings (or paragraph-topic sentences). For a 1,000–2,000 word synthesis brief, this is usually three to six sections, each tied to a sub-question of your main question.

4. The evidence map. Under each section, list the two to four sourced claims that will appear there, with a source tag for each. “Section 2 — Scope and Sequence. Claim 2a: Beast Academy’s published scope-and-sequence covers the 4th-grade CCSS topics but leaves gaps at 7th grade [source: Beast Academy publisher page, primary]. Claim 2b: parent reviews corroborate…” This is where the brief’s integrity is built. If a section has no sourced claims under it, that section will become fact-free connector prose.

5. The uncertainty ledger. A short list of things we are not confident about — disagreements between sources, single-source claims, areas the triangulation left open. These do not get papered over. They get named in the brief, usually near where the relevant claim appears. Most briefs need at least one; none should have more than three or four before the author asks whether the question can actually be answered.

6. What the reader does next. One to three sentences naming the decision or action the brief enables. If the brief is for a term paper, this is the role this material plays in the paper. If the brief is for a decision, it is the decision being made and the recommended direction.

The synthesis plan is what you hand the agent. The brief is what the agent drafts against the plan. You write the plan; the agent writes the prose; you read the prose line by line.

Holding uncertainty in the text CORE

A real research brief has uncertainty in it. The question is where the uncertainty lives.

Hedged postscript (wrong): a brief that states claims confidently throughout, then adds a paragraph at the end: “Of course, there are limitations to this analysis, including…” This is a fig leaf. Readers remember the confident prose, not the postscript.

In the text, near the claim (right): the uncertainty appears where the claim appears. “Thompson (2024) reports a modest negative effect; Garcia & Lee (2024) report no effect. The methodological difference is that Thompson controlled for household income and Garcia & Lee did not. For our purposes, we should treat the evidence as suggestive but contested.” A reader who reads only that paragraph still gets the honest picture.

The confidence-tier vocabulary from Lesson 4.3 helps. Low / medium / high confidence is a short, honest way to mark claims without sounding stilted. A synthesis brief with three or four claims marked with explicit confidence is more trustworthy than one that pretends to uniform certainty, and it is easier to read than one that sounds anxious throughout.

One useful move: write the brief twice if you are on the borderline. First, write it as if every claim were certain — to see what your most confident possible brief looks like. Then, revise to insert the uncertainty exactly where it lives. The gap between those two drafts is what your synthesis plan’s uncertainty ledger was for.

What research agents do well in synthesis, and what they do not CORE

An honest map, because a student who misunderstands this wastes effort.

Well: drafting prose from a synthesis plan. Given a good plan with source tags, the agent produces a readable first draft of a brief faster than a human would, and usually with better paragraph-level structure. Use the agent for this. Do not draft prose from scratch — it is not where your time should go.

Well: catching inconsistencies between your plan and your draft. “You said the thesis is X; paragraph 4 argues toward Y.” A good prompt to run on a draft: “Compare the thesis sentence of this brief to the actual argumentative direction of the body. Flag any paragraphs that drift from the thesis.” The agent is good at this.

Well: generating alternative phrasings, titles, and section headings. Low-stakes, easy to reject, useful when you are stuck on how to name something.

Poorly: holding uncertainty. Agents pull toward confident prose. You will have to re-insert the uncertainty in every draft, in multiple places. This is the single biggest synthesis workload that stays with you.

Poorly: the thesis. The thesis is the verdict you reached from your triangulation. The agent does not have the verdict — you do. If you ask the agent to “write the thesis,” you get a bland thesis that is technically supported by the sources and does not actually say anything. Write your own thesis.

Poorly: the uncertainty ledger. The agent will under-report uncertainty by default. You must push back explicitly — “list every claim in this brief that rests on a single source” — and re-insert.

The pattern: the agent is strong on prose-craft and weak on what is being said. That is most of the reason you still write the plan.